Research On Fluid Property Identification Method Of Pyroclastic Conglomerate Reservoir Based On Machine Learning

Due to the increasing exploration level of oil and gas fields and the decreasing amount of conventional oil and gas resources,pyroclastic reservoir is another important exploration field after sandstone,carbonate and other reservoirs.The pore types of Tongbomiao Formation in Hailaer Basin are mainly intergranular dissolution pores,intra granular dissolution pores and cast film pores,etc.According to the classification standards of reservoir porosity and permeability,the reservoir of Tongbomiao Formation belongs to the medium and low porosity-low ultra-low permeability reservoir.Pyroclastic rocks are the rocks formed by transporting and depositing volcanic lava through rivers,which contain both volcanic rocks,normal sedimentary rocks and transitional lithology.The parent rocks consist of neutral,basic,acidic and normal sedimentary rocks.The irregular distribution of different granulosity conglomerate causes the complex pore structure of pyroclastic conglomerate reservoir.Moreover,the same porosity of low-porosity and ultra-low permeability reservoirs has different permeability values.The tuffaceous and argillaceous filling has additional conductivity,which enhances the electrical conductivity of rocks.There are many factors affecting the resistivity curve,which brings certain difficulty to the logging interpretation.When using machine learning method as a means of fluid property identification,it can not only extract the nonlinear relationship between the data from the huge data body,but also greatly reduce the labor cost,so that the fluid property identification method of machine learning has obvious advantages over the manual interpretation of logging data.For the volcanic clastic conglomerate reservoir,conventional logging,NMR logging,laboratory NMR,core analysis,thin section and oil test data have been collected in the WB area in the early stage,and the logging interpretation and evaluation methods have been studied from the aspects of lithology identification,reservoir type classification,the relationship between electrical properties and pore space,fluid property identification model and so on.Finally,according to the characteristics of the mother rock of Tongbomiao Formation in WB area,the mother rock type identification chart of conventional logging data is established.Based on the scale division of laboratory NMR and NMR logging,the reservoir types in the study area were divided into three types corresponding to different productivity.According to the conductive properties of tuff and clay in the study area,the relevant characteristics of different pore Spaces and electrical properties were obtained respectively.By using the neural network method,the resistivity of saturated pure water in formation is predicted,and the logging data which is more sensitive to fluid property identification is selected.Combined with logging data,the fluid property identification of pyroclastic conglomerate reservoir in the study area was carried out by machine learning method,and verified by other test Wells.The machine learning architecture method with high coincidence rate was obtained.It provides a set of effective methods for lithology identification,pore space and electric relationship,reservoir classification and fluid property identification of pyroclastic glutenite in WB area,and provides a solid basis for logging interpretation for the exploration and development of pyroclastic glutenite reservoir in WB area.